Effect of general health checks on the treatment of chronic diseases: accounting for self-selection in the retrospective cohort study using Korea National Health Insurance data

Abstract

Objective This study examines the effect of general health checks on the detection and treatment of diabetes and hypertension with controlling for the self-selection problem of undergoing general health checks. Design Retrospective observational cohort study. Setting Sample Research Database offered by Korean National Health Insurance Service, between 2002 and 2013. Participants Two datasets, focusing on diabetes and hypertensions one by one, are constructed. The number of participants for the datasets is 133329 (diabetes) and 101738 (hypertension), respectively. Methods A bivariate probit model with selection was adopted to investigate the impact of general health checks on the diagnosis of critical chronic diseases. The dependent variable was an indicator variable denoting whether a participant has been treated for diabetes (or hypertension) or not for the first time during the sample period. An indicator variable that indicates whether that participant is eligible for free general health checks or not in the focal year (year of the first treatment or last year in the sample) was used as instrument variables to control for the self-selection problem of undergoing general health checks. Results We found that there exists substantial self-selection between undergoing general health checks and diagnosis for chronic diseases. The correlations between the unobserved factors influencing the decisions to obtain general health checks and those determining the detection of chronic diseases are highly significant and positive (ie, 0.188 (p<0.001) in diabetes and 0.220 (p<0.001) in hypertension). We confirmed that these positive, significant correlations generate upward bias in the estimated effect of general health checks on the detection and treatment of diabetes (0.312 (p<0.001) when self-selection ignored but 0.099 (p<0.001) when self-selection considered) and hypertension (0.293 (p<0.001) when self-selection ignored but insignificant when self-selection considered). The effect of general health checks and people's self-selection behaviour may differ by socio-economic characteristics of individuals. The general health check is effective in detecting chronic diseases among low-income individuals rather than high-income individuals, implying that general health checks are contributing to helping medically underprivileged low-income people detect and treat their chronic diseases. High-income individuals showed stronger self-selection behaviour than low-income individuals and this may overstate the effect of general health checks if the self-selection is overlooked, particularly among high-income individuals. Conclusion Self-selection due to unobserved factors between undergoing general health checks and diagnosis of chronic diseases are substantial. After accounting for this, the effect of general health checks on the detection and treatment of diabetes and hypertension is insignificant or marginal. The increases in the treatments of the two diseases following general health checks are 1% and insignificant in diabetes and hypertension, respectively.